Grok Word Limit by Model (2026)
xAI's Grok jumped from a 128k context ceiling to 2 million tokens in just over a year. Here's what fits in each current model, what it costs, and where the practical limits sit.
Quick Answer
Grok 4.1 accepts 2,000,000 tokens (~1.5M words), matching Gemini 2.5 Pro. Grok 4 accepts 256,000 tokens (~193K words). Grok 4 Fast and Grok 4.1 Fast also extend to 2M. Grok 4.20, released in beta February 2026, defaults to 200k with optional long-context mode up to 2M. Output is capped at roughly 16,384 tokens for most variants.
Grok context windows by model
| Model | Input tokens | Input words | Released |
|---|---|---|---|
| Grok 4.1 | 2,000,000 | ~1,500,000 | Nov 2025 |
| Grok 4.1 Fast | 2,000,000 | ~1,500,000 | Nov 2025 |
| Grok 4.20 Heavy | 200,000 (up to 2M) | ~150,000 – 1,500,000 | Feb 2026 |
| Grok 4 | 256,000 | ~193,000 | Jul 2025 |
| Grok 4 Fast | 2,000,000 | ~1,500,000 | Sep 2025 |
| Grok 3 (legacy) | 131,072 | ~97,000 | Feb 2025 |
Specs from xAI official documentation and OpenRouter model cards as of April 2026. Verify current specs with xAI before relying on these for production.
How Grok caught up on context length
Grok started behind. Grok 1 (March 2024) had 8k tokens of context at a time when GPT-4 Turbo was already at 128k. Grok 3 expanded to 131k. Grok 4, released July 2025, jumped to 256k. And then in November 2025, Grok 4.1 hit 2M tokens, which puts it in the same tier as Gemini 2.5 Pro and above everything OpenAI and Anthropic offer.
That 15-month trajectory from 8k to 2M is the steepest context expansion any major AI lab has ever shipped. It matters because context length is one of the few hard capability limits that is genuinely visible to end users. An engineer deciding whether to use Grok vs Claude vs Gemini for analyzing a large codebase cares about whether the whole thing fits in one prompt more than about which model has a slightly higher benchmark score.
What 2 million tokens actually holds
Grok 4.1's 2M token window is large enough to ingest things most AI models can't touch:
- The complete seven-book Harry Potter series (~1.08M words) with room left over
- War and Peace twice (~1.17M words combined)
- The King James Bible (~783K words) with a full-length novel on top
- A medium-sized application's full codebase (100K-500K lines of code)
- Several years of meeting transcripts at standard podcast word density
- Roughly 6,000 single-spaced pages
Whether all 2M tokens are genuinely usable for complex reasoning is a separate question. xAI's own marketing claims Grok 4.1 maintains answer quality through the full window. Independent evaluations tend to show some degradation past 500K-1M tokens on synthesis tasks (where the model has to reason across the entire context), though retrieval-style tasks (finding specific facts) hold up better. This pattern matches what research groups have found with every long-context model, not just Grok.
Pricing
Grok's pricing tiers at April 2026 list rates:
| Model | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
| Grok 4 | $3.00 | $15.00 |
| Grok 4 Fast | $0.20 | $0.50 |
| Grok 4 (above 128k tokens) | Higher tier | Contact xAI |
Two notable things. First, Grok 4 Fast at $0.20 per million input tokens is one of the cheapest 2M-context models available. Second, Grok 4 applies a surcharge on requests above 128k, which makes long-context Grok 4 runs more expensive than the spec sheet implies. The Fast variants don't have this cliff.
Consumer Grok on X vs the API
Grok is available three ways: inside the X (Twitter) app, through the Grok website, and via the xAI API. Limits differ across all three.
- X app (free): Grok 4 access with strict daily message caps. Context appears capped well below the API maximum.
- SuperGrok ($30/month): Unlimited Grok 4.20 access, longer context per message, no hard caps.
- SuperGrok Heavy ($300/month): 16-agent mode, priority access, enterprise-style availability.
- API: Full specs listed in the tables above. Pay per token. Suitable for production workloads.
If you're testing whether your 100k-word document fits, the X app will reject you long before the API would. Use the API or a tool like OpenRouter for real benchmarks.
Check your prompt against every model
See how your text fits in Grok, ChatGPT, Claude, Gemini, and more, side by side
AI Prompt Word CounterFAQ
What is Grok's word limit?
Grok 4.1 accepts about 1.5 million words (2M tokens). Grok 4 accepts about 193,000 words (256k tokens). Grok 4 Fast also handles 2M tokens for much lower cost.
Is Grok's 2M token window real or marketing?
It is real for retrieval tasks, where Grok holds up well deep into the window. For complex multi-step reasoning across the full 2M, some degradation past 500K-1M tokens is expected, matching the pattern seen with every long-context model.
Can Grok read entire book series?
Yes on Grok 4.1 and Grok 4 Fast. All seven Harry Potter books fit with room to spare. Lord of the Rings plus The Hobbit fits. War and Peace fits twice.
How much does it cost to fill Grok 4's context?
At $3 per million input tokens, filling Grok 4's 256k window is roughly $0.77. Above 128k, xAI applies a surcharge, so actual long-context costs are higher. Grok 4 Fast at $0.20/M is dramatically cheaper.
How does Grok compare to Claude and GPT on context?
Grok 4.1 (2M) matches Gemini 2.5 Pro and beats Claude (200K) and GPT-4.1 (1M) on raw context size. Quality within that context depends on the task and is disputed across benchmarks.